Add missing parameters to the `nan<x>` functions

Rel. #6198.

NumPy 1.22 added missing parameters for some nan<x> functions.

A number of the nan<x> functions previously lacked parameters that were present in their <x>-based counterpart, e.g. the where parameter was present in numpy.mean but absent from numpy.nanmean.

https://numpy.org/devdocs/release/1.22.0-notes.html#missing-parameters-have-been-added-to-the-nan-x-functions

• nanmin: initial & where
• nanmax: initial & where
• nanargmin: keepdims & out
• nanargmax: keepdims & out
• nansum: initial & where
• nanprod: initial & where
• nanmean: where
• nanvar: where
• nanstd: where

[takagi]

how would you pass the identity variable to the create_reduction_func

We have two options:

• Change the reduction kernel to take identity at runtime
• Create another reduction kernel instance when it takes initial other than the default

I think the latter is acceptable.

Also, keepdims is already a supported argument, just undocumented.

👍

As for where, is it just a simple change of using cupy.where as I believe create_reduction_func does not support where argument just yet?

It currently doesn’t support it. We may use cupy.where together, but it needs extra memory to have the result of where. cupy.ufunc internally supports _where keyword parameter, and its code may help you.

https://github.com/cupy/cupy/blob/9395dbe22ddb591df6e3d044c8b6877fcbbf7b99/cupy/_core/_kernel.pyx#L973-L978

[takagi]

Please go ahead as no progress for months, @praburocking!